PARTIAL LEAST SQUARES (PLS-SEM) 2016 Edition

Copyright @c 2016 by G. David Garson and Statistical Associates Publishing Page 1

 PARTIAL LEAST SQUARES (PLS-SEM) 2016 Edition

THIS IS THE PREVIEW VERSION OF THE FIRST 25 PAGES.

SEE BOTTOM FOR MORE INFORMATION ON THE FULL VERSION.

@c 2014, 2015, 2016 by G. David Garson and Statistical Associates Publishing. All rights
reserved worldwide in all media. No permission is granted to any user to copy or post this work
in any format or any media unless permission is granted in writing by G. David Garson and
Statistical Associates Publishing.

ISBN-10: 1626380392
ISBN-13: 978-1-62638-039-4
The author and publisher of this eBook and accompanying materials make no representation or
warranties with respect to the accuracy, applicability, fitness, or completeness of the contents
of this eBook or accompanying materials. The author and publisher disclaim any warranties
(express or implied), merchantability, or fitness for any particular purpose. The author and
publisher shall in no event be held liable to any party for any direct, indirect, punitive, special,
incidental or other consequential damages arising directly or indirectly from any use of this
material, which is provided as is, and without warranties. Further, the author and publisher
do not warrant the performance, effectiveness or applicability of any sites listed or linked to in
this eBook or accompanying materials. All links are for information purposes only and are not
warranted for content, accuracy or any other implied or explicit purpose. This eBook and
accompanying materials is copyrighted by G. David Garson and Statistical Associates
Publishing. No part of this may be copied, or changed in any format, sold, or rented in any way
under any circumstances, including selling or renting for free.

Contact:

G. David Garson, President

Statistical Publishing Associates
274 Glenn Drive
Asheboro, NC 27205 USA

Email: [email protected]
Web: www.statisticalassociates.com

Copyright @c 2016 by G. David Garson and Statistical Associates Publishing Page 2

 PARTIAL LEAST SQUARES (PLS-SEM) 2016 Edition

Table of Contents
Overview ......................................................................................................................................... 8
Data ................................................................................................................................................. 9
Key Concepts and Terms............................................................................................................... 10
Background............................................................................................................................... 10
Models ...................................................................................................................................... 13
Overview .............................................................................................................................. 13
PLS-regression vs. PLS-SEM models .................................................................................... 13
Components vs. common factors ........................................................................................ 14
Components vs. summation scales ..................................................................................... 16
PLS-DA models ..................................................................................................................... 16
Mixed methods .................................................................................................................... 16
Bootstrap estimates of significance .................................................................................... 17
Reflective vs. formative models .......................................................................................... 17
Confirmatory vs. exploratory models .................................................................................. 20
Inner (structural) model vs. outer (measurement) model .................................................. 21
Endogenous vs. exogenous latent variables ....................................................................... 21
Mediating variables ............................................................................................................. 22
Moderating variables........................................................................................................... 23
Interaction terms ................................................................................................................. 25
Partitioning direct, indirect, and total effects ....................... Error! Bookmark not defined.
Variables ...................................................................................... Error! Bookmark not defined.
Case identifier variable ........................................................... Error! Bookmark not defined.
Measured factors and covariates ........................................... Error! Bookmark not defined.
Modeled factors and response variables ............................... Error! Bookmark not defined.
Single-item measures ............................................................. Error! Bookmark not defined.
Measurement level of variables ............................................. Error! Bookmark not defined.
Parameter estimates ................................................................... Error! Bookmark not defined.
Cross-validation and goodness-of-fit .......................................... Error! Bookmark not defined.
PRESS and optimal number of dimensions ............................ Error! Bookmark not defined.
PLS-SEM in SPSS, SAS, and Stata ...................................................... Error! Bookmark not defined.
Overview ..................................................................................... Error! Bookmark not defined.
PLS-SEM in SmartPLS ....................................................................... Error! Bookmark not defined.
Overview ..................................................................................... Error! Bookmark not defined.
Estimation options in SmartPLS .................................................. Error! Bookmark not defined.
Running the PLS algorithm .......................................................... Error! Bookmark not defined.
Options ................................................................................... Error! Bookmark not defined.
Data input and standardization .............................................. Error! Bookmark not defined.
Setting the default workspace................................................ Error! Bookmark not defined.
Creating a PLS project and importing data............................. Error! Bookmark not defined.
Validating the data settings .................................................... Error! Bookmark not defined.
Drawing the path model ......................................................... Error! Bookmark not defined.
Reflective vs. formative models ............................................. Error! Bookmark not defined.

Copyright @c 2016 by G. David Garson and Statistical Associates Publishing Page 3

 PARTIAL LEAST SQUARES (PLS-SEM) 2016 Edition

Displaying/hiding the measurement model ........................... Error! Bookmark not defined.

Saving the model .................................................................... Error! Bookmark not defined.
Model report output .............................................................. Error! Bookmark not defined.
Checking for convergence ...................................................... Error! Bookmark not defined.
OUTPUT .................................................................................. Error! Bookmark not defined.
Path coefficients for the inner model..................................... Error! Bookmark not defined.
Direct, indirect, and total path coefficients ........................... Error! Bookmark not defined.
Outer model measurement loadings and weights ................. Error! Bookmark not defined.
Bootstrapped significance output .......................................... Error! Bookmark not defined.
Assessing model fit: Overview ................................................ Error! Bookmark not defined.
Measurement fit for reflective models .................................. Error! Bookmark not defined.
Measurement fit for formative models.................................. Error! Bookmark not defined.
Goodness of fit for structural models .................................... Error! Bookmark not defined.
Latent variable correlations output........................................ Error! Bookmark not defined.
Analyzing residuals ................................................................. Error! Bookmark not defined.
Estimation with the consistent PLS (PLSc) algorithm.................. Error! Bookmark not defined.
Overview ................................................................................. Error! Bookmark not defined.
PLSc output ............................................................................. Error! Bookmark not defined.
Estimation with PLS bootstrapping ............................................. Error! Bookmark not defined.
Overview ................................................................................. Error! Bookmark not defined.
Running the PLS bootstrapping algorithm ............................. Error! Bookmark not defined.
PLS bootstrap output .............................................................. Error! Bookmark not defined.
Dropping indicators ................................................................ Error! Bookmark not defined.
Estimation with consistent PLS bootstrapping ........................... Error! Bookmark not defined.
Overview ................................................................................. Error! Bookmark not defined.
Running the PLSc bootstrapping algorithm ............................ Error! Bookmark not defined.
Consistent PLS bootstrap output ............................................ Error! Bookmark not defined.
Dropping indicators ................................................................ Error! Bookmark not defined.
Estimation with blindfolding ....................................................... Error! Bookmark not defined.
Overview ................................................................................. Error! Bookmark not defined.
Running the blindfolding algorithm........................................ Error! Bookmark not defined.
Output unique to blindfolding estimation ............................. Error! Bookmark not defined.
Confirmatory tetrad analysis (CTA) ............................................. Error! Bookmark not defined.
Overview ................................................................................. Error! Bookmark not defined.
The example model ................................................................ Error! Bookmark not defined.
Running confirmatory tetrad analysis .................................... Error! Bookmark not defined.
PLS-CTA output ....................................................................... Error! Bookmark not defined.
PLS-CTA and sample size ........................................................ Error! Bookmark not defined.
Importance-performance map analysis (IPMA) .......................... Error! Bookmark not defined.
Overview ................................................................................. Error! Bookmark not defined.
The example model ................................................................ Error! Bookmark not defined.
Running IPMA ......................................................................... Error! Bookmark not defined.
Overview ................................................................................. Error! Bookmark not defined.

Copyright @c 2016 by G. David Garson and Statistical Associates Publishing Page 4

 PARTIAL LEAST SQUARES (PLS-SEM) 2016 Edition

IPMA Output ........................................................................... Error! Bookmark not defined.

Finite-mixture segmentation analysis (FIMIX) ............................ Error! Bookmark not defined.
Unobserved heterogeneity..................................................... Error! Bookmark not defined.
Comparing models with differing numbers of segments ....... Error! Bookmark not defined.
Fit Indices ................................................................................ Error! Bookmark not defined.
Entropy ................................................................................... Error! Bookmark not defined.
Path coefficients ..................................................................... Error! Bookmark not defined.
T-tests of differences in path coefficients .............................. Error! Bookmark not defined.
Labeling the segments ............................................................ Error! Bookmark not defined.
Prediction-oriented segmentation (POS) .................................... Error! Bookmark not defined.
Overview ................................................................................. Error! Bookmark not defined.
The example model ................................................................ Error! Bookmark not defined.
Running POS ........................................................................... Error! Bookmark not defined.
POS output.............................................................................. Error! Bookmark not defined.
Labeling the segments ............................................................ Error! Bookmark not defined.
Multi- group Analysis (MGA) ....................................................... Error! Bookmark not defined.
Overview ................................................................................. Error! Bookmark not defined.
Measurement invariance........................................................ Error! Bookmark not defined.
The example model ................................................................ Error! Bookmark not defined.
Defining groups....................................................................... Error! Bookmark not defined.
Running MGA.......................................................................... Error! Bookmark not defined.
Multi-group output ................................................................. Error! Bookmark not defined.
Testing for segment difference .............................................. Error! Bookmark not defined.
Permutation algorithm (MICOM) ................................................ Error! Bookmark not defined.
Overview ................................................................................. Error! Bookmark not defined.
The example model ................................................................ Error! Bookmark not defined.
Running the permutation algorithm ...................................... Error! Bookmark not defined.
Permutation algorithm output ............................................... Error! Bookmark not defined.
Measurement invariance (MICOM) tests ............................... Error! Bookmark not defined.
Type I and Type II error........................................................... Error! Bookmark not defined.
PLS regression modeling with SmartPLS .......................................... Error! Bookmark not defined.
PLS regression: SmartPLS vs SPSS or SAS .................................... Error! Bookmark not defined.
PLS regression: SPSS vs. SAS ........................................................ Error! Bookmark not defined.
Example ....................................................................................... Error! Bookmark not defined.
Creating a simple regression model in SmartPLS........................ Error! Bookmark not defined.
SmartPLS output for PLS-regression ........................................... Error! Bookmark not defined.
Path coefficient ....................................................................... Error! Bookmark not defined.
Outer loadings and weights .................................................... Error! Bookmark not defined.
Model fit/quality criteria ........................................................ Error! Bookmark not defined.
Other output ........................................................................... Error! Bookmark not defined.
PLS regression modeling with SPSS ................................................. Error! Bookmark not defined.
Overview ..................................................................................... Error! Bookmark not defined.
SPSS example............................................................................... Error! Bookmark not defined.

Copyright @c 2016 by G. David Garson and Statistical Associates Publishing Page 5

 PARTIAL LEAST SQUARES (PLS-SEM) 2016 Edition

SPSS input .................................................................................... Error! Bookmark not defined.

SPSS output ................................................................................. Error! Bookmark not defined.
Proportion of variance explained by latent factors................ Error! Bookmark not defined.
PRESS (predicted error sum of squares) ................................. Error! Bookmark not defined.
Latent component weights and loadings ............................... Error! Bookmark not defined.
Variable importance in the projection (VIP) for the independent variablesError! Bookmark
not defined.
Regression parameter estimates by dependent variable ...... Error! Bookmark not defined.
Charts/plots ............................................................................ Error! Bookmark not defined.
Plots of latent factor weights ................................................. Error! Bookmark not defined.
Residual and normal quantile plots ........................................ Error! Bookmark not defined.
Saving variables ...................................................................... Error! Bookmark not defined.
PLS regression modeling using SAS.................................................. Error! Bookmark not defined.
Overview ..................................................................................... Error! Bookmark not defined.
SAS example ................................................................................ Error! Bookmark not defined.
SAS syntax ................................................................................... Error! Bookmark not defined.
SAS output ................................................................................... Error! Bookmark not defined.
R-square analysis .................................................................... Error! Bookmark not defined.
Correlation loading plot.......................................................... Error! Bookmark not defined.
Variable Importance Plot ........................................................ Error! Bookmark not defined.
Response scores by predictor scores plots ............................ Error! Bookmark not defined.
Residuals plots ........................................................................ Error! Bookmark not defined.
Parameter estimates .............................................................. Error! Bookmark not defined.
Summary ................................................................................. Error! Bookmark not defined.
PLS regression modeling using Stata ............................................... Error! Bookmark not defined.
Overview ..................................................................................... Error! Bookmark not defined.
Assumptions..................................................................................... Error! Bookmark not defined.
Robustness .................................................................................. Error! Bookmark not defined.
Parametric v. non-parametric ..................................................... Error! Bookmark not defined.
Independent observations .......................................................... Error! Bookmark not defined.
Data level ..................................................................................... Error! Bookmark not defined.
Unobserved homogeneity ........................................................... Error! Bookmark not defined.
Linearity ....................................................................................... Error! Bookmark not defined.
Outliers ........................................................................................ Error! Bookmark not defined.
Residuals...................................................................................... Error! Bookmark not defined.
Appropriate sample size.............................................................. Error! Bookmark not defined.
Missing values ............................................................................. Error! Bookmark not defined.
Model specification ..................................................................... Error! Bookmark not defined.
Appropriate model fit assessment .............................................. Error! Bookmark not defined.
Recursivity ................................................................................... Error! Bookmark not defined.
Multicollinearity .......................................................................... Error! Bookmark not defined.
Proper use of dummy variables .................................................. Error! Bookmark not defined.
Standardized variables ................................................................ Error! Bookmark not defined.

Copyright @c 2016 by G. David Garson and Statistical Associates Publishing Page 6

 PARTIAL LEAST SQUARES (PLS-SEM) 2016 Edition

Frequently Asked Questions ............................................................ Error! Bookmark not defined.

How does PLS-SEM compare to SEM using analysis of covariance structures? ................. Error!
Bookmark not defined.
What are higher order/hierarchical component models in PLS-SEM? ...... Error! Bookmark not
defined.
How does norming work? ........................................................... Error! Bookmark not defined.
What other software packages support PLS-regression? ........... Error! Bookmark not defined.
How is PLS installed for SPSS? ..................................................... Error! Bookmark not defined.
What other software packages support PLS-SEM apart from SmartPLS? . Error! Bookmark not
defined.
What are the SIMPLS and PCR methods in PROC PLS in SAS? .... Error! Bookmark not defined.
Why is PLS sometimes described as a 'soft modeling' technique?............ Error! Bookmark not
defined.
You said PLS could handle large numbers of independent variables, but can't OLS regression
do this too? ................................................................................. Error! Bookmark not defined.
Is PLS always a linear technique? ................................................ Error! Bookmark not defined.
How is PLS related to principal components regression (PCR) and maximum redundancy
analysis (MRA)? ........................................................................... Error! Bookmark not defined.
What are the NIPALS and SVD algorithms? ................................ Error! Bookmark not defined.
How does PLS relate to two-stage least squares (2SLS)?............ Error! Bookmark not defined.
How does PLS relate to neural network analysis (NNA)? ........... Error! Bookmark not defined.
Acknowledgments............................................................................ Error! Bookmark not defined.
Bibliography ..................................................................................... Error! Bookmark not defined.

Copyright @c 2016 by G. David Garson and Statistical Associates Publishing Page 7

 PARTIAL LEAST SQUARES (PLS-SEM) 2016 Edition

Partial Least Squares

Overview
Partial least squares (PLS) analysis is an alternative to OLS regression, canonical
correlation, or covariance-based structural equation modeling (SEM) of systems
of independent and response variables. In fact, PLS is sometimes called
composite-based SEM, "component-based SEM", or variance-based SEM, in
contrast to "covariance-based SEM," which is the usual type (e.g., implemented
by Amos, SAS, Stata, MPlus, LISREL, EQS and other major software packages).
On the response side, PLS can relate the set of independent variables to multiple
dependent (response) variables. On the predictor side, PLS can handle many
independent variables, even when predictors display multicollinearity. PLS may be
implemented as a regression model, predicting one or more dependents from a
set of one or more independents; or it can be implemented as a path model,
handling causal paths relating predictors as well as paths relating the predictors to
the response variable(s). PLS is implemented as a regression model by SPSS and
by SAS's PROC PLS. SmartPLS is the most prevalent implementation as a path
model.

PLS is characterized as a technique most suitable where the research purpose is

prediction or exploratory modeling. In general, covariance-based SEM is preferred
when the research purpose is confirmatory modeling. PLS is less than satisfactory
as an explanatory technique because it is low in power to filter out variables of
minor causal importance (Tobias, 1997: 1).

The advantages of PLS include ability to model multiple dependents as well as

multiple independents; ability to handle multicollinearity among the
independents; robustness in the face of data noise and missing data; and creating
independent latent variables directly on the basis of cross-products involving the
response variable(s), making for stronger predictions. Disadvantages of PLS
include greater difficulty of interpreting the loadings of the independent latent
variables (which are based on cross-product relations with the response variables,
not based as in common factor analysis on covariances among the manifest
independents) and because the distributional properties of estimates are not

Copyright @c 2016 by G. David Garson and Statistical Associates Publishing Page 8

 PARTIAL LEAST SQUARES (PLS-SEM) 2016 Edition

known, the researcher cannot assess significance except through bootstrap

induction.

Overall, the mix of advantages and disadvantages means PLS is favored as a

predictive technique and not as an interpretive technique, except for exploratory
analysis as a prelude to an interpretive technique such as multiple linear
regression or covariance-based structural equation modeling. Hinseler, Ringle,
and Sinkovics (2009: 282) thus state, "PLS path modeling is recommended in an
early stage of theoretical development in order to test and validate exploratory
models."

Developed by Herman Wold (Wold, 1975, 1981, 1985) for econometrics and
chemometrics and extended by Jan-Bernd Lohmller (1989) , PLS has since spread
to research in education (ex., Campbell & Yates, 2011), marketing (ex., Albers,
2009, cites PLS as the method of choice in success factors marketing research),
and the social sciences (ex., Jacobs et al., 2011). See Lohmller (1989) for a
mathematical presentation of the path modeling variant of PLS, which compares
PLS with OLS regression, principal components factor analysis, canonical
correlation, and structural equation modeling with LISREL.

Data
Data for the section on PLS-SEM with SmartPLS uses the file jobsat.csv, a comma-
delimited file which may also be read by many other statistical packages. For the
jobsat file, sample size is 932. All variables are metric. Data are fictional and used
for instructional purposes only. Variables in the jobsat.* file included these:
StdEduc: respondents educational level, standardized
OccStat: respondents occupational status
Motive1: Score on the Motive1 motivational scale
Motive2: Score on the Motive2 motivational scale
Incent1: Score on the Incent1 incentives scale
Incent2: Score on the incent2 incentives scale
Gender: Coded 0=Male, 1=Female. Used for PLS-MGA (multigroup analysis)
SPSS, SAS, and Stata versions of jobsat.* are available below.
Click here for jobsat.csv, for SmartPLS

Copyright @c 2016 by G. David Garson and Statistical Associates Publishing Page 9

 PARTIAL LEAST SQUARES (PLS-SEM) 2016 Edition

Click here for jobsat4.csv, for PLS-CTA in SmartPLS

Click here for jobsat.sav, for SPSS
Click here for jobsat.sas7bdat, for SAS
Click here for jobsat.dta, for Stata
The section of PLS regression modeling in SmartPLS uses the HappyLife data
file.
Click here for HappyLife.csv, for SmartPLS
The section on PLS regression modeling in SPSS uses the SPSS sample file, 1991
U.S. General Social Survey.sav. The USGSS1991.sas7bdat file is a different format
of the same variables used for the SAS example.
Click here for 1991 U.S. General Social Survey.sav, for SPSS
Click here for USGSS1991.sas7bdat, for SAS

Key Concepts and Terms

Background
Partial least squares (PLS) is sometimes called "Projection to Latent Structures"
because of its general strategy, which is illustrated below. Note, however, that
there may be more than one X component and more than one Y component and
the arrows connecting the components to their indicators may be reflective as
shown or may be reversed (reflective vs. formative modeling is discussed below).

Copyright @c 2016 by G. David Garson and Statistical Associates Publishing Page 10

 PARTIAL LEAST SQUARES (PLS-SEM) 2016 Edition

The steps in the PLS-SEM algorithm, as described by Henseler, Ringle, & Sarstedt
(2012), are summarized below.

1. Preliminary to applying the PLS algorithm, the measured indicator variables

are normalized to have a mean of 0 and a standard deviation of 1. In this
context, normalized means standardized. PLS requires standardized latent
variable scores, and since latent variables in PLS are linear combinations of
the indicator variables, it is necessary that the indicator variables be
standardized. A consequence is that both measurement (outer model) and
structural (inner model) path coefficients vary from 0 to plus or minus 1,
with paths closest to absolute 1 being the strongest

2. In the first stage of the PLS algorithm, the measured indicator variables are
used to create the X and Y component scores. To do this, an iterative
process is used, looping repeatedly through four steps:

i. Latent variable scores are given initial approximations based on

equally weighted indicator scores.
ii. Initial weights are assigned to the structural (inner) paths connecting
the latent variables using a path weighting scheme based on
regression, to maximize the R-squared of each endogenous latent
variable. That is, component scores as estimated in a given iteration
are used to calculate structural path weights. Put a third way, using

Copyright @c 2016 by G. David Garson and Statistical Associates Publishing Page 11

 PARTIAL LEAST SQUARES (PLS-SEM) 2016 Edition

regression, successive iterations adjust the structural weights to

maximize the strength of the relation of successive pairs of X and Y
component scores by maximizing the covariance of each X-score with
the Y variables. This maximizes the explained variance of the
dependent component.
* Note that alternative to this path weighting scheme, factor-
based and centroid-based weighting schemes may be used by
different algorithms. PLS-regression in SPSS and SAS, for
instance, uses a factor-based weighting scheme. See below.
iii. The structural (inner) weights are used to adjust the estimates of the
latent variable scores.
iv. The measurement (outer) weights connecting the latent variables to
their indicator variables are estimated differently, depending on
whether the model is reflective or formative (that is, which way the
measurement arrows go, discussed below). For the usual reflective
model, with arrows going from the latent variable to the indicator
variables, measurement path weights are based on the covariances
between the estimate of the latent variable and the indicator
variable. If the model is formative, with arrows going from the
indicators to the latent variable, measurement path weights are
based on regression of the latent variable on its indicator.

Iterations of these four steps stop when there is no significant change in the
measurement (outer) weights of the indicator variables. The weights of the
indictor variables in the final iteration are the basis for computing the final
estimates of latent variable scores. The final latent variable scores, in turn, are
used as the basis of OLS regressions to calculate the final structural (inner)
weights in the model.

The overall result of the PLS algorithm is that the components of X are used to
predict the scores of the Y components, and the predicted Y component scores
are used to predict the actual values of the measured Y variables. This strategy
means that while the original X variables may be multicollinear, the X components
used to predict Y will be orthogonal. Also, the X variables may have missing
values, but there will be a computed score for every case on every X component.
Finally, since only a few components (often two or three) will be used in
predictions, PLS coefficients may be computed even when there may have been

Copyright @c 2016 by G. David Garson and Statistical Associates Publishing Page 12

 PARTIAL LEAST SQUARES (PLS-SEM) 2016 Edition

more original X variables than observations (though results are more reliable with
more cases). In contrast, any of these three conditions (multicollinearity, missing
values, and too few cases in relation to number of variables) may well render
traditional OLS regression estimates unreliable or impossible. The same is true of
estimates by other procedures in the general and generalized linear model
families.

Models
Overview
Partial least squares as originally developed in the 1960s by Wold was a general
method which supported modeling causal paths among any number of "blocks" of
variables (latent variables), somewhat akin to covariance-based structural
equation modeling, the subject of the separate Statistical Associates "Blue Book"
volume, Structural Equation Modeling. PLS-regression models are a subset of
PLS-SEM models, where there are only two blocks of variables: the independent
block and the dependent block. SPSS and SAS implement PLS-regression models.
For more complex path models it is necessary to employ specialized PLS software.
SmartPLS is perhaps the most popular and is the software used in illustrations
below, but there are alternatives discussed below.

PLS-regression vs. PLS-SEM models

PLS-regression models are an alternative to OLS regression or canonical
correlation. PLS regression has been used, for instance, for econometric growth
modeling (Korkmazoglu & Kemalbay, 2012). Comparing OLS and PLS regression,
one simulation study (Temme, Kreis, & Lutz, 2006: 20) found "In our study, results
for simulated data, however, are very similar to those resulting from OLS
regression."

PLS-SEM models, in contrast, are path models in which some variables may be
effects of others while still be causes for variables later in the hypothesized causal
sequence. PLS-SEM models are an alternative to covariance-based structural
equation modeling (traditional SEM).

For an extended critique of PLS-SEM as not being suitable as a structural equation

modeling (SEM) approach, see Rnkk & Evermann (2013). For critique of Rnkk

Copyright @c 2016 by G. David Garson and Statistical Associates Publishing Page 13

 PARTIAL LEAST SQUARES (PLS-SEM) 2016 Edition

& Evermann, see Henseler, Dijkstra, Sarstedt et. al. (2014). To summarize this
complex exchange in simple terms, Rnkk & Evermann took the view that PLS
yields inconsistent and biased estimates compared to traditional SEM, and in
addition PLS-SEM lacks a test for over-identification. Defending PLS, Henseler,
Dijkstra, Sarstedt et. al. took the view that Rnkk & Evermann wrongly assumed
SEM must revolve around common factors and failed to recognize that structural
equation models allow more general measurement models than traditional factor
analytic structures on which traditional SEM is based (Bollen & Long, 1993: 1).
That is, these authors argued that PLS should be seen as a more general form of
SEM, supportive of composite as well as common factor models (for an opposing
view, see McIntosh, Edwards, & Antonakis, 2014). Henseler et al. wrote scholars
have started questioning the reflex-like application of common factor models
(Rigdon, 2013). A key reason for this skepticism is the overwhelming empirical
evidence indicating that the common factor model rarely holds in applied
research (as noted very early by Schnemann & Wang, 1972). For example,
among 72 articles published during 2012 in what Atinc, Simmering, and Kroll
(2012) consider the four leading management journals that tested one or more
common factor model(s), fewer than 10% contained a common factor model that
did not have to be rejected.

The critical bottom line for the researcher, agreed upon by both sides of the
composite vs. common factors debate, is that factors do not have the same
meaning in PLS-SEM models as they do in traditional SEM models. Coefficients
from the former therefore do not necessarily correspond closely to those from
the latter. As in all statistical approaches, it is not a matter of a technique being
right or wrong but rather it is a matter of properly understanding what the
technique is.

Components vs. common factors

As mentioned above, traditional variance-based PLS-SEM is a component-based
approach using a type of principle components analysis to construct latent
variables. This contrasts with traditional covariance-based SEM, which uses a type
of common factor analysis to create latent variables. While latent variables are
traditionally conceptualized as common factors, a construct might also be a
component, which is what they are in PLS-SEM.

Copyright @c 2016 by G. David Garson and Statistical Associates Publishing Page 14

 PARTIAL LEAST SQUARES (PLS-SEM) 2016 Edition

Common factors models assume that all the covariation among the set of its
indicator variables is explained by the common factor. In a pure common factor
model in SEM, in graphical/path terms, arrows are drawn from the factor to the
indicators, and neither direct arrows nor covariance arrows connect the
indicators.

Component factors, also called composite factors, have a more general model of
the relationship of indicators to factors. Specifically, it is not assumed that all
covariation among the set of indicator variables is explained by the factor. Rather,
covariation may also be explained by relationships among the indicators. In
graphical/path terms, covariance arrows may connect each indicator with each
other indicator in its set. Henseler, Dijkstra, Sarstedt, et al. (2014: 185) note, the
composite factor model does not impose any restrictions on the covariances
between indicators of the same construct.

Traditional common factor based SEM seeks to explain the covariance matrix,
including covariances among the indicators. Specifically, in the pure common
factor model, the model-implied covariance matrix assumes covariances relating
indicators within its own set or with those in other sets is 0 (as reflected by the
absence of connecting arrows in the path diagram). Goodness of fit is typically
assessed in terms of the closeness of the actual and model-implied covariance
matrices.

Traditional component based PLS-SEM does not attempt to explain the

covariances connecting indicators. Henseler, Dijkstra, Sarstedt, et al. (2014: 186)
write, The composite factor model leaves the covariation between indicators of
the same block unexplained, which means that the implied covariances between
these indicators equal the empirical covariances.

In summary, latent variables in traditional PLS-SEM are components whereas in

traditional SEM they are common factors. Because of this, PLS-SEM can be said to
be more general than traditional SEM. Traditional SEM can be said to be more
parsimonious than traditional PLS-SEM (fewer arrows in its model). Whether the
researchers data fits a common factor model or fits a component model better is
an empirical question, the answer to which is conditional on the particular data at
hand. Advocates of composite factor models are likely to assert that component
(composite) models are more prevalent in the real world (Bentler & Huang, in
press). In practice, researchers using traditional SEM may add covariance arrows
Copyright @c 2016 by G. David Garson and Statistical Associates Publishing Page 15
 PARTIAL LEAST SQUARES (PLS-SEM) 2016 Edition

connecting indicators precisely because good fit cannot be achieved without

moving toward a composite factor model (Bentler, 1976).

This discussion applies to the traditional PLS algorithm. The consistent PLS
(PLSc) algorithm discussed further below is employed in conjunction with
common factor models and not with composite models. For a discussion of
components vs. common factors in modeling, see Henseler, Dijkstra, Sarstedt, et
al. (2014). For criticism, see McIntosh, Edwards, & Antonakis, 2014).

Components vs. summation scales

It is common in social science to create scales, which are a type of latent variable,
simply by making the scale an additive (or otherwise weighted) sum of the
indicators. In the PLS world, this is called the sum score approach. Typically, the
sum score approach weights each indicator equally. The PLS-SEM approach, in
contrast, gives more weight to indicators with higher predictive validity.

As noted by Henseler, Dijkstra, Sarstedt, et al. (2014: 192), PLS construct scores
can only be better than sum scores if the indicators vary in terms of the strength
of relationship with their underlying construct. If they do not vary, any method
that assumes equally weighted indicators will outperform PLS. That is, PLS-SEM
assumes that indicators vary in the degree that each is related to the measured
latent variable. If not, summation scales are preferable. SmartPLS 3 will use the
sum scores approach if, as discussed below, maximum iterations are set to 0.

PLS-DA models
PLS-DA models are PLS discriminant analysis models. These are an alternative to
discriminant function analysis, for PLS regression models where the
dependent/response variable is binary variable or a dummy variable rather than a
block of continuous variables.

Mixed methods
Note that researchers may combine both PLS regression modeling with PLS-SEM
modeling. For instance, Tenenhaus et al. (2004), in a marketing study, used PLS
regression to obtain a graphical display of products and their characteristics, with
a mapping of consumer preferences. Then PLS-SEM was used to obtain a detailed

Copyright @c 2016 by G. David Garson and Statistical Associates Publishing Page 16

 PARTIAL LEAST SQUARES (PLS-SEM) 2016 Edition

analysis of each consumer group by building a causal model involving consumer

preference, physico-chemical, and sensory blocks of variables.

Bootstrap estimates of significance

As the distribution of PLS is unknown, conventional significance testing is
impossible. However, testing may be accomplished by resampling methods such
as the bootstrap or the jacknife, as illustrated by Davies (2001). Resampling
methods do not have specific sample size requirements but the smaller the
sample, the more likely that fitted confidence limits will be fitted to noise in the
data rather than to a true underlying distribution. Wakeling & Morris (2005),
based on Monte Carlo simulation studies, have created tables of critical values of
r2cv. Since resampling estimates are data-driven, results may not hold up for other
datasets and therefore cross-validation is recommended (ex., developing the PLS
model for even-numbered observations an validating it for odd-numbered
observations, after randomizing the order of observations).

Some PLS packages use bootstrapping (e.g., SmartPLS) while other use jackknifing
(e.g., PLS-GUI). Both result in estimates of the standard error of regression paths
and other model parameters. The estimates are usually very similar.
Bootstrapping, which involves taking random samples and randomly replacing
dropped values, will give slightly different standard error estimates on each run.
Jackknifing, which involves a leave-one-out approach for n 1 samples, will
always give the same standard error estimates. Where jackknifing estimates the
point variance, bootstrapping estimates the point variance and the entire
distribution and thus bootstrapping is required when the research purpose is
distribution estimation. As the research purpose is much more commonly
variance estimation, jackknifing is often preferred on grounds of replicability and
being less computationally intensive.

Reflective vs. formative models

A path model is reflective if in the path diagram causal arrows go from the latent
variable (factor) to the measured indicator variables. A path model is formative if
the arrows go from the observed measures to the latent variables. Sometimes
reflective models are called Mode A models and formative models are called
Mode B models. SmartPLS allows both types of models (choose Selection, Invert
measurement model from the menu).
Copyright @c 2016 by G. David Garson and Statistical Associates Publishing Page 17
 PARTIAL LEAST SQUARES (PLS-SEM) 2016 Edition

Both traditional SEM and PLS-SEM support both reflective and formative models.
By historical tradition, reflective models have been the norm in structural
equation modeling and formative models have been the norm in partial least
squares modeling. This is changing as researchers become aware that the choice
between reflective and formative models should depend on the nature of the
indicators.

It should be noted there is a controversy in the literature with regard to the

proper statistical modeling procedure for reflective and formative models.
Adherents to the PLS approach, such as Christian Ringle, Oliver Gtz, et al. (2009).
(2014) and Jrg Henseler, Theo Dijkstra, Marko Sarstedt, et al. (2014), see PLS
being applicable to both reflective and formative models. Critics such as
McIntosh, Edwards, & Antonakis (2014: 215) express the view that PLS should be
applied only to formative composite models (see discussion above), not to
common factors models, since traditional SEM (covariance-based SEM) has
superior statistical properties for reflective latent variable models.

In reflective models, indicators are a representative set of items which all reflect
the latent variable they are measuring. Reflective models assume the factor is the
"reality" and measured variables are a sample of all possible indicators of that
reality. This implies that dropping one indicator may not matter much since the

Copyright @c 2016 by G. David Garson and Statistical Associates Publishing Page 18

 PARTIAL LEAST SQUARES (PLS-SEM) 2016 Edition

other indicators are representative also. The latent variable will still have the
same meaning after dropping one indicator.

In formative models, each indicator represents a dimension of meaning of the

latent variable. The set of indicators collectively represent all dimensions of the
latent variable. Formative models assume the indicators are "reality" and are all
the dimensions of the factor. Dropping an indicator in a formative model is
equivalent to dropping a dimension of meaning, causing the meaning of the latent
variable to change. To the extent that a dimension dropped is important, the
meaning of the latent variable will change. The researcher may retain the same
label for the latent variable (ex., motivation) even though the dimensions
change, misleading readers into thinking motivation means the same thing
across models.

Albers and Hildebrandt (2006; sourced in Albers, 2010: 412) give an example for a
latent variable dealing with satisfaction with hotel accommodations. A reflective
model might have the representative measures I feel well in this hotel, This
hotel belongs to my favorites, I recommend this hotel to others, and I am
always happy to stay overnight in this hotel. A formative model, in contrast,
might have the constituent measures, The room is well equipped, I can find
silence here, The fitness area is good, The personnel are friendly, and The
service is good.

Suppression is another possible problem in formative models. Because the

indicator items in a formative model represent different dimensions, there is the
possibility one item may be negatively correlated with another. If a first item is
negatively related to a second indicator but positively related to the latent
variable, the positive correlation of the second indicator with the latent variable
may be suppressed due push-pull effect of the co-occurrence of positive and
negative weights (see Henseler, Ringle, & Sarstedt, 2012: 270).

SmartPLS outputs a table of Indicator Data (Correlations), shown below. Output

for reflective and formative models is identical. Coefficients are the raw bivariate
correlations. For a reflective model, one would expect the correlations of
indicators for the same construct would be high since indicators should be
representative measures of the same phenomenon. For instance, the correlation
of Incent1 with Incent2 for the construct Incentives is .656. For a formative
model, indicators should represent different dimensions of the phenomenon and
Copyright @c 2016 by G. David Garson and Statistical Associates Publishing Page 19
 PARTIAL LEAST SQUARES (PLS-SEM) 2016 Edition

would not be expected to necessarily correlate highly unless there are multiple
measures for the same dimension. Examination of this table provides one type of
evidence of whether the data should be modeled reflectively or formatively.

In terms of estimation methods discussed further below, the traditional PLS

algorithm is well suited for formative and composite models. In early applications
of PLS, formative modeling was more prevalent than reflective modeling, quite in
contrast to the tradition of covariance-based structural equation modeling, where
reflective modeling has always been dominant. Some, such as Rigdon (2013), have
even argued that the use of PLS should be restricted to formative and composite
measurement models (see response to Rigdon by Dijkstra, 2014). The consistent
PLS algorithm (PLSc), discussed below, is designed to produce consistent
estimated for reflective models and may be preferred in reflective applications.

Confirmatory vs. exploratory models

Most statisticians see PLS is seen as an exploratory procedure. Wold himself
(1981) advised against using PLS for confirmatory models since PLS lacks
goodness-of-fit tests, which in turn means absence of criteria for selecting the
optimum model. The literature, however, reveals more confirmatory than
exploratory uses of PLS modeling by a large margin.

Copyright @c 2016 by G. David Garson and Statistical Associates Publishing Page 20

 PARTIAL LEAST SQUARES (PLS-SEM) 2016 Edition

Inner (structural) model vs. outer (measurement) model

When a model is created, factors are represented by ellipses and indicator
(measured) variables by rectangles. The factor ellipses and arrow creating them
are called the inner or structural model. The indicator rectangles and arrows
connecting them are called the outer or measurement model. This is illustrated
below.

Endogenous vs. exogenous latent variables

These are terms pertaining to the inner or structural model. A latent variable is
exogenous if it is not an effect of any other latent variable in the model (there are
no incoming arrows from other latent variables). A latent variable is endogenous
it is an effect of at least one other latent variable (there is at least one incoming
arrow from another latent variable). In the diagram above, SES and INCENTIVES
are exogenous while MOTIVATION is endogenous.

Copyright @c 2016 by G. David Garson and Statistical Associates Publishing Page 21

 PARTIAL LEAST SQUARES (PLS-SEM) 2016 Edition

Mediating variables
A mediating variable is simply an intervening variable. In the model below,
Motivation is a mediating variable between SES and Incentives on the one hand
and Productivity on the other.
If there were also direct paths from SES and/or Incentives to Productivity, the SES
and/or Incentives would be anteceding variables (or moderating variables as
defined below) for both Motivation and Productivity. Motivation would still be a
mediating variable.
A common type of mediator analysis involving just such mediating and
moderating effects is to start with a direct path, say SES -> Productivity, then see
what the consequences are when an indirect, mediated path is added, such as SES
-> Motivation -> Productivity. There are a number of possible findings when the
mediated path is added:
The correlation of SES and Productivity drops to 0, meaning there is no SES-
>Productivity path as the entire causality is mediated by Motivation. This is
called a full control effect of Motivation as a mediating variable.
The correlation of SES and Productivity remains unchanged, meaning
mediated path is inconsequential. This is no effect.
The correlation of SES and Productivity drops only part way toward 0,
meaning both the direct and indirect paths exist. This is called partial
control by the mediating variable.
The correlation of SES and Productivity increases compared to the original,
unmediated model. This is called suppression and would occur in this
example if the effect of SES directly on productivity and the effect of SES
directly on Motivation were opposite in sign, creating a push-pull effect.

Copyright @c 2016 by G. David Garson and Statistical Associates Publishing Page 22

 PARTIAL LEAST SQUARES (PLS-SEM) 2016 Edition

Moderating variables
The term moderating variable has been used in different and sometimes
conflicting ways by various authors. Some writers use mediating and
moderating interchangeably or flip the definition compared to other scholars.
For instance, a mediating variable as described above does affect or moderate
the relationship of variables it separates in a causal chain and thus might be called
a moderating variable. It is also possible to model interactions between latent
variables and latent variables representing interactions may be considered to
involve moderating variables. Multigroup analysis of heterogeneity across groups,
discussed further below, is also a type of analysis of a moderating effect.
However, as used here, a moderating variable is an anteceding joint direct or
indirect cause of two variables further down in the causal model. In the
illustration below, SES is modeled as an anteceding cause of both Incentives and
Motivation.

Copyright @c 2016 by G. David Garson and Statistical Associates Publishing Page 23

 PARTIAL LEAST SQUARES (PLS-SEM) 2016 Edition

In the model above, adding SES to the model may cause the path from Incentives
to Motivation to remain the same (no effect), drop to 0 (complete control effect
of SES), drop part way to 0 (partial control effect), or to increase (suppression
effect).
Spurious effects
If two variables share an anteceding cause, they usually are correlated but this
effect may be spurious. That is, it may be an artifact of mutual causation. A classic
example is ice cream sales and fires. These are correlated, but when the
anteceding mutual cause of heat of the day is added, the original correlation
goes away. In the model above, similarly, if the original correlation of Incentives
and Motivation disappeared when the mutual anteceding cause SES was added to
the model, it could be inferred that the originally observed effect of Incentives on
Motivation was spurious.
Suppression
A suppression effect occurs when the anteceding variable is positively related to
the predictor variable (ex., Incentives) and negatively related to the effect
variable (ex., Motivation). In such a situation the anteceding variable has a
suppressing effect in that the original Incentives/Motivation correlation without
SES in the model will be lower than when SES is added to the model, revealing its
push-pull effect as an anteceding variable. Put another way, the effect of

Copyright @c 2016 by G. David Garson and Statistical Associates Publishing Page 24

 PARTIAL LEAST SQUARES (PLS-SEM) 2016 Edition

Incentives on Motivation initially appears weaker than it really is because, as seen

when SES is added to the model, the correlation is suppressed by SES as a
moderating variable.

Interaction terms
An interaction term is an exogenous moderator variable which affects an
endogenous variable by way of a non-additive joint relation with another
exogenous variable. While it may also have a direct effect on the endogenous
target variable, the interaction is its non-additive joint relationship In the diagram
below, top, without interaction effect, latent variables A and B are modeled as
causes of Y. Let moderator variable M be added as a third cause of Y. The
researcher may suspect, however, that M and A have a joint effect on Y which
goes beyond the separate A and M linear effects that is, an interaction effect is
suspected.
There are two popular methods for modeling such a hypothesized interaction.
The first, the product indicator method, is illustrated below. This method may
only be used for reflective models. In this approach, a new latent variable (the
A*M factor) is added to the model whose indicators are the products of every
possible pair of indicators for A and for M. For instance, its first indicator is
INDM1*INDA1, being the product of the first indicator for M times the first
indicator for A. If there is an interaction effect beyond the separate linear effects
of A and M, then the path from A*M to Y will be significant.

END OF PREVIEW OF FIRST 25 PAGES

The full book has 237 more pages.
To buy the Kindle version for $7.00, click here.
To buy the entire Statistical Associates library of 50 statistics titles in no-password
pdf format on DVD plus one year of free updates for $120, click here.
To register for a password-protected pdf version when available, go to
http://www.statisticalassociates.com

Copyright @c 2016 by G. David Garson and Statistical Associates Publishing Page 25

 PARTIAL LEAST SQUARES (PLS-SEM) 2016 Edition

PARTIAL LEAST SQUARES REGRESSION AND STRUCTURAL EQUATION MODELS

Overview

This is a graduate-level introduction and illustrated tutorial on partial least

squares (PLS). PLS may be used in the context of variance-based structural
equation modeling, in contrast to the usual covariance-based structural equation
modeling, or in the context of implementing regression models. PLS is largely a
nonparametric approach to modeling, not assuming normal distributions in the
data, often recommended when the focus of research is prediction rather than
hypothesis testing, when sample size is not large, or in the presence of noisy data.

Why we think the new edition is important:

Covers the much-enhanced Version 3 of SmartPLS

Now book length at 262 pages (was 139)
Covers the traditional PLS algorithm and consistent PLS
Covers bootstrapped PLS, consistent bootstrapped PLS, and PLS with
blindfolding
Covers confirmatory tetrad analysis
Covers importance-performance map analysis (IPMA)
Covers finite-mixture segmentation (FIMIX) and prediction-oriented
segmentation (POS)
Covers multi-group analysis (MGA)
Covers significance testing with the permutation algorithm (MICOM)

Below is the unformatted table of contents.

PARTIAL LEAST SQUARES: REGRESSION AND STRUCTURAL EQUATION MODELS

Overview 8
Data 9
Key Concepts and Terms 10
Background 10
Models 13
Overview 13
PLS-regression vs. PLS-SEM models 13
Components vs. common factors 14
Components vs. summation scales 16
PLS-DA models 16
Mixed methods 16
Bootstrap estimates of significance 17

Copyright @c 2016 by G. David Garson and Statistical Associates Publishing Page 26

 PARTIAL LEAST SQUARES (PLS-SEM) 2016 Edition

Reflective vs. formative models 17

Confirmatory vs. exploratory models 20
Inner (structural) model vs. outer (measurement) model 21
Endogenous vs. exogenous latent variables 21
Mediating variables 22
Moderating variables 23
Interaction terms 25
Partitioning direct, indirect, and total effects 28
Variables 29
Case identifier variable 30
Measured factors and covariates 30
Modeled factors and response variables 30
Single-item measures 31
Measurement level of variables 32
Parameter estimates 33
Cross-validation and goodness-of-fit 33
PRESS and optimal number of dimensions 34
PLS-SEM in SPSS, SAS, and Stata 35
Overview 35
PLS-SEM in SmartPLS 35
Overview 35
Estimation options in SmartPLS 36
Running the PLS algorithm 37
Options 37
Data input and standardization 40
Setting the default workspace 41
Creating a PLS project and importing data 41
Validating the data settings 43
Drawing the path model 44
Reflective vs. formative models 47
Displaying/hiding the measurement model 48
Saving the model 49
Model report output 52
Checking for convergence 57
OUTPUT 58
Path coefficients for the inner model 58
Direct, indirect, and total path coefficients 59
Outer model measurement loadings and weights 60
Bootstrapped significance output 62
Assessing model fit: Overview 62
Measurement fit for reflective models 63
Measurement fit for formative models 73
Goodness of fit for structural models 79
Latent variable correlations output 85
Analyzing residuals 86
Estimation with the consistent PLS (PLSc) algorithm 88
Overview 88
PLSc output 90
Estimation with PLS bootstrapping 91
Overview 91
Running the PLS bootstrapping algorithm 92
PLS bootstrap output 97
Dropping indicators 103
Estimation with consistent PLS bootstrapping 104
Overview 104
Running the PLSc bootstrapping algorithm 104

Copyright @c 2016 by G. David Garson and Statistical Associates Publishing Page 27

 PARTIAL LEAST SQUARES (PLS-SEM) 2016 Edition

Consistent PLS bootstrap output 109

Dropping indicators 115
Estimation with blindfolding 115
Overview 115
Running the blindfolding algorithm 116
Output unique to blindfolding estimation 117
Confirmatory tetrad analysis (CTA) 122
Overview 122
The example model 123
Running confirmatory tetrad analysis 124
PLS-CTA output 125
PLS-CTA and sample size 128
Importance-performance map analysis (IPMA) 128
Overview 128
The example model 129
Running IPMA 131
Overview 133
IPMA Output 134
Finite-mixture segmentation analysis (FIMIX) 137
Unobserved heterogeneity 137
Comparing models with differing numbers of segments 139
Fit Indices 143
Entropy 147
Path coefficients 149
T-tests of differences in path coefficients 152
Labeling the segments 152
Prediction-oriented segmentation (POS) 154
Overview 154
The example model 156
Running POS 156
POS output 158
Labeling the segments 166
Multi- group Analysis (MGA) 166
Overview 166
Measurement invariance 166
The example model 167
Defining groups 168
Running MGA 172
Multi-group output 175
Testing for segment difference 180
Permutation algorithm (MICOM) 182
Overview 182
The example model 183
Running the permutation algorithm 183
Permutation algorithm output 184
Measurement invariance (MICOM) tests 185
Type I and Type II error 187
PLS regression modeling with SmartPLS 189
PLS regression: SmartPLS vs SPSS or SAS 189
PLS regression: SPSS vs. SAS 189
Example 190
Creating a simple regression model in SmartPLS 190
SmartPLS output for PLS-regression 192
Path coefficient 192
Outer loadings and weights 193
Model fit/quality criteria 194

Copyright @c 2016 by G. David Garson and Statistical Associates Publishing Page 28

 PARTIAL LEAST SQUARES (PLS-SEM) 2016 Edition

Other output 196

PLS regression modeling with SPSS 197
Overview 197
SPSS example 197
SPSS input 199
SPSS output 202
Proportion of variance explained by latent factors 202
PRESS (predicted error sum of squares) 202
Latent component weights and loadings 203
Variable importance in the projection (VIP) for the independent variables
205
Regression parameter estimates by dependent variable 206
Charts/plots 207
Plots of latent factor weights 210
Residual and normal quantile plots 211
Saving variables 211
PLS regression modeling using SAS 212
Overview 212
SAS example 213
SAS syntax 213
SAS output 214
R-square analysis 214
Correlation loading plot 215
Variable Importance Plot 216
Response scores by predictor scores plots 217
Residuals plots 218
Parameter estimates 220
Summary 221
PLS regression modeling using Stata 221
Overview 221
Assumptions 222
Robustness 222
Parametric v. non-parametric 223
Independent observations 224
Data level 225
Unobserved homogeneity 225
Linearity 226
Outliers 226
Residuals 226
Appropriate sample size 226
Missing values 228
Model specification 229
Appropriate model fit assessment 229
Recursivity 230
Multicollinearity 230
Proper use of dummy variables 231
Standardized variables 231
Frequently Asked Questions 231
How does PLS-SEM compare to SEM using analysis of covariance structures?
231
What are higher order/hierarchical component models in PLS-SEM? 235
How does norming work? 236
What other software packages support PLS-regression? 237
How is PLS installed for SPSS? 238
What other software packages support PLS-SEM apart from SmartPLS? 239
What are the SIMPLS and PCR methods in PROC PLS in SAS? 240

Copyright @c 2016 by G. David Garson and Statistical Associates Publishing Page 29

 PARTIAL LEAST SQUARES (PLS-SEM) 2016 Edition

Why is PLS sometimes described as a 'soft modeling' technique? 241

You said PLS could handle large numbers of independent variables, but can't
OLS regression do this too? 241
Is PLS always a linear technique? 241
How is PLS related to principal components regression (PCR) and maximum
redundancy analysis (MRA)? 241
What are the NIPALS and SVD algorithms? 243
How does PLS relate to two-stage least squares (2SLS)? 243
How does PLS relate to neural network analysis (NNA)? 243
Acknowledgments 244
Bibliography 245
Pagecount: 262

Copyright 2008, 2014, 2015, 2016 by G. David Garson and Statistical Associates Publishers.
Worldwide rights reserved in all media. Do not post on other servers, even for educational use.

Copyright @c 2016 by G. David Garson and Statistical Associates Publishing Page 30

 PARTIAL LEAST SQUARES (PLS-SEM) 2016 Edition

Statistical Associates Publishing

Blue Book Series
NEW! For use by a single individual, our entire current library is available at Amazon in no-password pdf format
on DVD for $120 plus shipping. Click on http://www.amazon.com/dp/1626380201 . Includes one year of free
updates when email address is provided.

Association, Measures of
Case Study Research
Cluster Analysis
Correlation
Correlation, Partial
Correspondence Analysis
Cox Regression
Creating Simulated Datasets
Crosstabulation
Curve Estimation & Nonlinear Regression
Data Management, Introduction to
Delphi Method in Quantitative Research
Discriminant Function Analysis
Ethnographic Research
Explaining Human Behavior
Factor Analysis
Focus Group Research
Game Theory
Generalized Linear Models/Generalized Estimating Equations
GLM Multivariate, MANOVA, and Canonical Correlation
GLM: Univariate & Repeated Measures
Grounded Theory
Life Tables & Kaplan-Meier Survival Analysis
Literature Review in Research and Dissertation Writing
Logistic Regression: Binary & Multinomial
Log-linear Models,
Longitudinal Analysis
Missing Values & Data Imputation
Multidimensional Scaling
Multilevel Linear Mixed Models
Multiple Regression
Narrative Analysis
Network Analysis
Neural Network Models

Copyright @c 2016 by G. David Garson and Statistical Associates Publishing Page 31

 PARTIAL LEAST SQUARES (PLS-SEM) 2016 Edition

Ordinal Regression
Parametric Survival Analysis
Partial Correlation
Partial Least Squares Regression & Structural Equation Models
Participant Observation
Path Analysis
Power Analysis
Probability
Probit and Logit Response Models
Propensity Score Analysis
Research Design
Scales and Measures
Significance Testing
Structural Equation Modeling
Survey Research & Sampling
Testing Statistical Assumptions
Two-Stage Least Squares Regression
Validity & Reliability
Variance Components Analysis
Weighted Least Squares Regression

Statistical Associates Publishing

http://www.statisticalassociates.com
[email protected]

Copyright @c 2016 by G. David Garson and Statistical Associates Publishing Page 32